Embedding
임베딩
Also known as: vector representation
A representation of text or images as a vector of numbers, where similar meanings sit closer together.
Compared letter by letter, cat and kitten look unrelated. An embedding moves meaning into coordinates so that close meanings land close together.
That changes search: you can find documents that mean the same thing without sharing a word. Recommendation, duplicate detection and classification rest on the same idea.
Images work the same way, and a vision language model puts both in one space so the distance between them can be measured.
- DimensionsTypically hundreds to thousands of numbers.
- CautionClose vectors do not mean equal facts. Verification is still needed.